Dual ISO - massive dynamic range improvement (dual_iso.mo)

[tron]

Solved indeed Many thanks.

[AdamTheBoy]

https://www.dropbox.com/s/d3n7qd18i1b653b/M232154J.000378.JPG

https://www.dropbox.com/s/20mhvdfwbnfmfp1/M232154J.000378.dng

https://www.dropbox.com/s/j60135pvsx09bb8/M232154J.000380.dng

Here's a raw video still processed using the latest cr2hdr!  I'd say it's looking pretty damn good, I wanted to try a particularly alias prone shot to see how well the new techniques were working.  I can't wait to see where this goes.

I also put up an unprocessed and processed DNG to check out if you'd like.

[swinxx]

can someone please post a link to a working conversion version for raw video, cause i can not make it work.. with that one i have.. and i can only find the newest one in the first post.
thx you so much. greets. sw

[ted ramasola]

@a1ex,

Heres a test comparing aug 16 vs aug 27 crshdr.



numbers correspond to numbered circles in attached image.

1. aug 27 better in handling fine frequency lines, better in handling aliasing and moire.

2. bright pixels popping out not shown in previous version

3. dark pixels/noise? popping out not present in previous.

Emailed to you cr2 of this image

[a1ex]

Yeah, these dark pixels are easily mistaken for aliasing (and viceversa). Not that easy to find a good balance between the two.

[ted ramasola]

Here is the same set up but I used a VAF filter.

Used aug 16 build.

Obviously the VAF plays a key physical component in this approach, but those without it can go for either versions?

this version offers control over hot pixels, + vaf cuts down moire and aliasing.

Is there a middle ground?

[a1ex]

For the troublesome image, here are the components that are being blended:

fullres - best detail, least aliasing, but noisy in shadows
fullres_smooth - the above, but with chroma filtered (used in highlights, where one ISO is overexposed)
halfres - least noise, but aliased; not used directly
halfres_smooth - the above, with chroma filtered; used in deep shadows if the algorithm thinks it's unlikely to have aliasing

The likelyhood of getting aliasing is computed from the difference between fullres and halfres images (and filtered a little).

In this image, the resolution chart detail is quite dark (close to the ISO 100 noise floor), so the algorithm chooses the less noisy version. Also, there's strong banding in the fullres shot (didn't see it that bad in the other shots). Also, the ISO 100 noise is a bit higher than normal (sensor got hot?)

At a closer look, the two exposures are not properly matched (ISO 100 shows a strong green cast in shadows), so the root of the problem is when matching the ISOs. The sensor response is probably nonlinear, and my algorithm assumes it's perfectly linear.

[Greg]

Will the 500D how to use Dual ISO?

You need to compile the module from the repository.


Canon dual iso without moire?
http://image-sensors-world.blogspot.se/2013/08/canon-files-for-dual-range-column.html

[a1ex]

If it has 20 stops of DR, I'm sold

I've tried to shoot a moonrise timelapse, and dual ISO struggled a bit. Didn't develop the shots yet, but I had to clip the moon in order to see the other stuff on the screen (not just pure black and street lamps).

@ted: I'm afraid the 5D2 has a nasty behavior in the optical black area. Here's the ISO 100 exposure after black correction:



The ISO 1600 exposure is fine, but it couldn't be matched with the ISO 100 one because of this.

[ted ramasola]

I think I prefer the half res smooth.

[a1ex]

I think I've solved it with some aggressive nonlinear correction. The two exposures are now adjusted at each bright raw level; before, I've used a gain (ISO) and an offset (black correction).

I've uploaded it to cr2hdr_exp.exe, since it suffers from overfitting in some cases (e.g. Danne's duck).

BTW, even if it looked like a hot pixel problem, I didn't touch that code at all.

[ted ramasola]

BOOM!!

You nailed it A1ex!

Clean shadow areas, very minimal aliasing and moire!

If that can be achieved in raw video, I'd throw away the VAF filter!!

[Danne]

Gotta try this out asap on some old files. Awesome!

[a1ex]

Danne's duck working again (a simple outlier check did the trick); all my pixel peeping tests passed (I have a script that runs it through 40 difficult test shots collected from the forum).

Also solved the problem from aseek (tested with 500 pics and still going strong).

=> updated cr2hdr from first post.

[Shutterlab Creative]

____URGENT____

Can someone PLEASE link me to where I can find the newest working download of CR2DNG? I was foolishly experimenting with this and then forgot to switch it off before a paid shoot. I have to process these photos tonight, Please help!

This will be for 5D3 CR2 files. I'm not sure if that makes a difference in which conversion version I will use.

Thanks for your help.

EDIT:

I've tried using the link to the CR2HDR.exe on the first page of this thread, but each time I drop the .CR2 files on the executable file command prompt opens very briefly and then nothing happens. What am I doing wrong? 

[Danne]

folder name, keep it without spaces

[Oedipax]

cr2hdr is running great and my shots look even better with this new version. However, exiftool doesn't seem to be working. It's in the same folder as cr2hdr, do I need to do something different?

Here's how it looks (Windows 7 64bit running in VMware on OS X):

[Shutterlab Creative]

I have named my folder with the copied CR2 files "processing" without any spaces. It's created a blank text file (tmp.txt) but I can't get these to process correctly. Does CR2HDR create DNG copies of the CR2 files automatically?

I was using the July 22nd build for the firmware just fyi.

[Oedipax]

Shutterlab, are you dragging your cr2's over from another folder? You have to have cr2hdr, dcraw and exiftool in the same folder as the cr2 files you're dragging onto it for it to work.

[Shutterlab Creative]

Ahhh. I was hoping it was something simple. Thanks for the tip!

[1%]

Also on win7 make sure to unblock all of these programs, cr2hdr can't run exif tool if its restricted.

[tron]

hello, I have noticed that in the specific CR2 I had sent although dark shadows do not show these repeatable patterns, I can see them in  mid tones (well actually less deep shadows). I have tested using the cr2hdr.exe from the first post and the _exp variance too.

(I must remind myself how the shadows would look with the Shadows slider close to maximum though in a normal Canon raw file though to keep things in perspective)

https://www.dropbox.com/s/7ilrw9bouph1t5u/cr2hdr_27aug_0708pm.JPG
https://www.dropbox.com/s/gvwfedjsujpx1rk/cr2hdr_28aug_1222am.JPG
https://www.dropbox.com/s/01osc2bj3ud8rka/cr2hdr_exp_28aug_0332am.JPG

I have uploaded the above screenshots (you have the .CR2) to see what I mean and for comparison...

[Oedipax]

Also on win7 make sure to unblock all of these programs, cr2hdr can't run exif tool if its restricted.

Not sure if I'm missing a step - I went into properties for exiftool just now and set "Full Control" for every account option I could find. I ran it as Administrator as well. Looking at that screenshot though it looks like it just can't find the file because something weird is happening with the name.

[ted ramasola]

Not sure if I'm missing a step - I went into properties for exiftool just now and set "Full Control" for every account option I could find. I ran it as Administrator as well. Looking at that screenshot though it looks like it just can't find the file because something weird is happening with the name.

try to move the folder containing all the cr2 and the cr2hdr files is in the root of your drive not in another folder.

[a1ex]

@tron: this time you are oversharpening the picture. I don't see anything bad in my jpeg.